#include <kernels.h>

static vx_status VX_CALLBACK validateGatherLayer(vx_node node, const vx_reference *parameters, vx_uint32 num, vx_meta_format metas[]) {
    vx_enum type, type2, out_type;
    vx_size num_dims, num_dims2, out_num_dims;
    vx_size input_dims[4], input_dims2[4], output_dims[4];

    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[0], VX_TENSOR_NUMBER_OF_DIMS, &num_dims, sizeof(num_dims)));
    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[0], VX_TENSOR_DATA_TYPE, &type, sizeof(type)));
    if ((type != VX_TYPE_FLOAT32) && (type != VX_TYPE_FLOAT16)) return VX_ERROR_INVALID_TYPE;
    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[0], VX_TENSOR_DIMS, input_dims, sizeof(input_dims)));

    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[1], VX_TENSOR_NUMBER_OF_DIMS, &num_dims2, sizeof(num_dims2)));
    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[1], VX_TENSOR_DATA_TYPE, &type2, sizeof(type2)));
    if ((type2 != VX_TYPE_INT32) && (type2 != VX_TYPE_INT64)) return VX_ERROR_INVALID_TYPE;
    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[1], VX_TENSOR_DIMS, input_dims2, sizeof(input_dims2)));

    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[2], VX_TENSOR_NUMBER_OF_DIMS, &out_num_dims, sizeof(out_num_dims)));
    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[2], VX_TENSOR_DATA_TYPE, &out_type, sizeof(out_type)));
    if ((out_type != VX_TYPE_FLOAT32) && (out_type != VX_TYPE_FLOAT16)) return VX_ERROR_INVALID_TYPE;
    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[2], VX_TENSOR_DIMS, output_dims, sizeof(output_dims)));

    // the output num_dim should equal to (input_dim + input_dim2 - 1)
    if (num_dims + num_dims2 - 1 != out_num_dims) {
        printf("validate: gather: The [rank(output tensor)] should equal to [rank(input tensor) + rank(indices tensor) - 1)]\n");
        printf("validate: gather: %d != %d + %d - 1\n", (int)out_num_dims, (int)num_dims, (int)num_dims2);
        return VX_ERROR_INVALID_DIMENSION;
    }
    
    vx_uint32 axis;
    ERROR_CHECK_STATUS(vxCopyScalar((vx_scalar)parameters[3], &axis, VX_READ_ONLY, VX_MEMORY_TYPE_HOST));
    if (axis < 0 || axis > 3) {
        printf("validate: gather: Axis value should be 0~2\n");
        printf("validate: gather: Axis = %d\n", axis);
        return VX_ERROR_INVALID_PARAMETERS;
    } 
    ERROR_CHECK_STATUS(vxSetMetaFormatAttribute(metas[2], VX_TENSOR_DATA_TYPE, &out_type, sizeof(out_type)));
    ERROR_CHECK_STATUS(vxSetMetaFormatAttribute(metas[2], VX_TENSOR_NUMBER_OF_DIMS, &out_num_dims, sizeof(num_dims)));
    ERROR_CHECK_STATUS(vxSetMetaFormatAttribute(metas[2], VX_TENSOR_DIMS, &output_dims, sizeof(output_dims)));

    return VX_SUCCESS;

}

static vx_status VX_CALLBACK query_target_support(vx_graph graph, vx_node node,
    vx_bool use_opencl_1_2,
    vx_uint32& supported_target_affinity
)
{
    supported_target_affinity = AGO_TARGET_AFFINITY_GPU;
    return VX_SUCCESS;
}

#if ENABLE_OPENCL
static vx_status VX_CALLBACK opencl_codegen(
    vx_node node,                                  // [input] node
    const vx_reference parameters[],               // [input] parameters
    vx_uint32 num,                                 // [input] number of parameters
    bool opencl_load_function,                     // [input]  false: normal OpenCL kernel; true: reserved
    char opencl_kernel_function_name[64],          // [output] kernel_name for clCreateKernel()
    std::string& opencl_kernel_code,               // [output] string for clCreateProgramWithSource()
    std::string& opencl_build_options,             // [output] options for clBuildProgram()
    vx_uint32& opencl_work_dim,                    // [output] work_dim for clEnqueueNDRangeKernel()
    vx_size opencl_global_work[],                  // [output] global_work[] for clEnqueueNDRangeKernel()
    vx_size opencl_local_work[],                   // [output] local_work[] for clEnqueueNDRangeKernel()
    vx_uint32& opencl_local_buffer_usage_mask,     // [output] reserved: must be ZERO
    vx_uint32& opencl_local_buffer_size_in_bytes   // [output] reserved: must be ZERO
)
{
    //get input params
    vx_size num_of_dims, num_of_dims2;
    vx_enum type;
    vx_uint32 axis;

    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[0], VX_TENSOR_NUMBER_OF_DIMS, &num_of_dims, sizeof(num_of_dims)));
    vx_size input_dims[num_of_dims];
    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[0], VX_TENSOR_DIMS, input_dims, sizeof(input_dims)));    

    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[1], VX_TENSOR_NUMBER_OF_DIMS, &num_of_dims2, sizeof(num_of_dims)));
    vx_size input_dims2[num_of_dims2];
    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[1], VX_TENSOR_DIMS, input_dims2, sizeof(input_dims2)));

    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[0], VX_TENSOR_DATA_TYPE, &type, sizeof(type)));
    ERROR_CHECK_STATUS(vxCopyScalar((vx_scalar)parameters[3], &axis, VX_READ_ONLY, VX_MEMORY_TYPE_HOST));
    
    strcpy(opencl_kernel_function_name, "gather_layer");

    //reverse input dims w,h,c,n- > n,c,h,w
    int start = 0, end = num_of_dims-1;
    while (start < end) {
        int temp = input_dims[start];
        input_dims[start] = input_dims[end];
        input_dims[end] = temp;
        start++;
        end--;
    }

    opencl_work_dim = 3;
    opencl_global_work[0] = 1;
    opencl_global_work[1] = 1;
    opencl_global_work[2] = 1;
    
    for (int i=0; i<num_of_dims; i++) {
        if (i < axis) {
            opencl_global_work[2] *= input_dims[i];
        }
        else if (i > axis) {
            opencl_global_work[0] *= input_dims[i];
        }
    }
    
    for (int i=0; i<num_of_dims2; i++) {
        opencl_global_work[1] *= input_dims2[i];
    }

    // Setting variables required by the interface
    opencl_local_buffer_usage_mask = 0;
    opencl_local_buffer_size_in_bytes = 0;

    if (num_of_dims) {
        char item[8192];
        if (type == VX_TYPE_FLOAT32) {
            sprintf(item,
                "#pragma OPENCL EXTENSION cl_amd_media_ops : enable\n"
                "__kernel void %s(__global uchar * in, uint in_offset, uint4 in_stride, __global uchar * ind, uint ind_offset, uint4 ind_stride, __global uchar * out, uint out_offset, uint4 out_stride, uint axis) \n"
                "{ \n"
                "   uint x = get_global_id(0);\n"
                "   uint y = get_global_id(1);\n"
                "   uint c = get_global_id(2);\n"
                "   int indices = *(__global int*)&ind[ind_offset + y*ind_stride.s0];\n"
                "   float value;\n"
                "   uint offset;\n"
                "   if (axis == 0) {\n"
                "       value = *(__global float*)&in[in_offset + x*in_stride.s0 + indices*in_stride.s1 + c*in_stride.s2];\n"
                "       offset = out_offset + x*out_stride.s0 + y*out_stride.s1 + c*out_stride.s2;\n"
                "   }\n"
                "   else if (axis == 1) {\n"
                "       value = *(__global float*)&in[in_offset + indices*in_stride.s0 + c*in_stride.s1];\n"
                "       offset = out_offset + y*out_stride.s0 + c*out_stride.s1;\n"
                "   }\n"
                "   else if (axis == 2) {\n"
                "       value = *(__global float*)&in[in_offset + c*in_stride.s0];\n"
                "       offset = out_offset + c*out_stride.s0;\n"
                "   }\n"
                "   out += offset;\n"
                "   *(__global float *)&out[0] = value;\n"
                "}\n", opencl_kernel_function_name);
        }
        else {
            sprintf(item,
                "#pragma OPENCL EXTENSION cl_amd_media_ops : enable\n"
                "__kernel void %s(__global uchar * in, uint in_offset, uint4 in_stride, __global uchar * ind, uint ind_offset, uint4 ind_stride, __global uchar * out, uint out_offset, uint4 out_stride, uint axis) \n"
                "{ \n"
                "   uint x = get_global_id(0);\n"
                "   uint y = get_global_id(1);\n"
                "   uint c = get_global_id(2);\n"
                "   int indices = *(__global int*)&ind[ind_offset + y*ind_stride.s0];\n"
                "   half value;\n"
                "   uint offset;\n"
                "   if (axis == 0) {\n"
                "       value = *(__global half*)&in[in_offset + x*in_stride.s0 + indices*in_stride.s1 + c*in_stride.s2];\n"
                "       offset = out_offset + x*out_stride.s0 + y*out_stride.s1 + c*out_stride.s2;\n"
                "   }\n"
                "   else if (axis == 1) {\n"
                "       value = *(__global half*)&in[in_offset + indices*in_stride.s0 + c*in_stride.s1];\n"
                "       offset = out_offset + y*out_stride.s0 + c*out_stride.s1;\n"
                "   }\n"
                "   else if (axis == 2) {\n"
                "       value = *(__global half*)&in[in_offset + c*in_stride.s0];\n"
                "       offset = out_offset + c*out_stride.s0;\n"
                "   }\n"
                "   out += offset;\n"
                "   *(__global half *)&out[0] = value;\n"
                "}\n", opencl_kernel_function_name);
        }
        opencl_kernel_code = item;
    }
    return VX_SUCCESS;
}
#endif

//! \brief The kernel execution.
static vx_status VX_CALLBACK host_kernel(vx_node node, const vx_reference * parameters, vx_uint32 num) 
{
#if ENABLE_HIP
    //get input params
    vx_size num_of_dims, num_of_dims2;
    vx_enum type;
    vx_uint32 axis;

    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[0], VX_TENSOR_NUMBER_OF_DIMS, &num_of_dims, sizeof(num_of_dims)));
    vx_size input_dims[num_of_dims];
    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[0], VX_TENSOR_DIMS, input_dims, sizeof(input_dims)));

    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[1], VX_TENSOR_NUMBER_OF_DIMS, &num_of_dims2, sizeof(num_of_dims)));
    vx_size input_dims2[num_of_dims2];
    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[1], VX_TENSOR_DIMS, input_dims2, sizeof(input_dims2)));

    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[0], VX_TENSOR_DATA_TYPE, &type, sizeof(type)));
    ERROR_CHECK_STATUS(vxCopyScalar((vx_scalar)parameters[3], &axis, VX_READ_ONLY, VX_MEMORY_TYPE_HOST));

    //reverse input dims w,h,c,n- > n,c,h,w
    std::reverse(input_dims, input_dims + num_of_dims);

    dim3 globalThreads = dim3(1);
    for (int i = 0; i < num_of_dims; i++) {
        if (i < axis) {
            globalThreads.z *= input_dims[i];
        }
        else if (i > axis) {
            globalThreads.x *= input_dims[i];
        }
    }
    for (int i = 0; i < num_of_dims2; i++) {
        globalThreads.y *= input_dims2[i];
    }

    vx_size temp[4] = {0};
    uint4 input_stride, ind_stride, output_stride;
    vx_size in_offset, ind_offset, output_offset;
    unsigned char *input_mem = NULL;
    unsigned char *ind_mem = NULL;
    unsigned char *output_mem = NULL;
    hipStream_t hip_stream;

    ERROR_CHECK_STATUS(vxQueryNode(node, VX_NODE_ATTRIBUTE_AMD_HIP_STREAM, &hip_stream, sizeof(hip_stream)));
    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[0], VX_TENSOR_BUFFER_HIP, &input_mem, sizeof(input_mem)));
    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[0], VX_TENSOR_OFFSET_GPU, &in_offset, sizeof(in_offset)));
    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[1], VX_TENSOR_BUFFER_HIP, &ind_mem, sizeof(ind_mem)));
    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[1], VX_TENSOR_OFFSET_GPU, &ind_offset, sizeof(ind_offset)));
    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[2], VX_TENSOR_BUFFER_HIP, &output_mem, sizeof(output_mem)));
    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[2], VX_TENSOR_OFFSET_GPU, &output_offset, sizeof(output_offset)));

    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[0], VX_TENSOR_STRIDE_GPU, temp, sizeof(temp)));
    input_stride.x = temp[0];
    input_stride.y = temp[1];
    input_stride.z = temp[2];
    input_stride.w = temp[3];

    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[1], VX_TENSOR_STRIDE_GPU, temp, sizeof(temp)));
    ind_stride.x = temp[0];
    ind_stride.y = temp[1];
    ind_stride.z = temp[2];
    ind_stride.w = temp[3];

    ERROR_CHECK_STATUS(vxQueryTensor((vx_tensor)parameters[2], VX_TENSOR_STRIDE_GPU, temp, sizeof(temp)));
    output_stride.x = temp[0];
    output_stride.y = temp[1];
    output_stride.z = temp[2];
    output_stride.w = temp[3];

    if (HipExec_Gather_layer(hip_stream, globalThreads, dim3(1), type, input_mem, in_offset, input_stride,
        ind_mem, ind_offset, ind_stride, output_mem, output_offset, output_stride, axis)) {
        return VX_FAILURE;
    }

    return VX_SUCCESS;

#elif ENABLE_OPENCL
    return VX_ERROR_NOT_IMPLEMENTED;
#endif
}

//! \brief The kernel publisher.
vx_status publishGatherLayer(vx_context context) 
{
    vx_kernel kernel = vxAddUserKernel(context, "com.amd.nn_extension.gather_layer", VX_KERNEL_GATHER_LAYER_AMD, host_kernel, 4, validateGatherLayer, nullptr, nullptr);
    ERROR_CHECK_OBJECT(kernel);

    amd_kernel_query_target_support_f query_target_support_f = query_target_support;
    ERROR_CHECK_STATUS(vxSetKernelAttribute(kernel, VX_KERNEL_ATTRIBUTE_AMD_QUERY_TARGET_SUPPORT, &query_target_support_f, sizeof(query_target_support_f)));

#if ENABLE_OPENCL
    amd_kernel_opencl_codegen_callback_f opencl_codegen_callback_f = opencl_codegen;
    ERROR_CHECK_STATUS(vxSetKernelAttribute(kernel, VX_KERNEL_ATTRIBUTE_AMD_OPENCL_CODEGEN_CALLBACK, &opencl_codegen_callback_f, sizeof(opencl_codegen_callback_f)));
#endif

    ERROR_CHECK_STATUS(vxAddParameterToKernel(kernel, 0, VX_INPUT, VX_TYPE_TENSOR, VX_PARAMETER_STATE_REQUIRED));
    ERROR_CHECK_STATUS(vxAddParameterToKernel(kernel, 1, VX_INPUT, VX_TYPE_TENSOR, VX_PARAMETER_STATE_REQUIRED));
    ERROR_CHECK_STATUS(vxAddParameterToKernel(kernel, 2, VX_OUTPUT, VX_TYPE_TENSOR, VX_PARAMETER_STATE_REQUIRED));
    ERROR_CHECK_STATUS(vxAddParameterToKernel(kernel, 3, VX_INPUT, VX_TYPE_SCALAR, VX_PARAMETER_STATE_REQUIRED));
    
    //finalize and release kernel object.
    ERROR_CHECK_STATUS(vxFinalizeKernel(kernel));
    ERROR_CHECK_STATUS(vxReleaseKernel(&kernel));

    return VX_SUCCESS; 
}

VX_API_ENTRY vx_node VX_API_CALL vxGatherLayer(vx_graph graph, vx_tensor input, vx_tensor indices, vx_tensor output, vx_int32 axis) 
{
    vx_node node = NULL;
    vx_context context = vxGetContext((vx_reference)graph);
    vx_scalar s_axis = vxCreateScalarWithSize(context, VX_TYPE_INT32, &axis, sizeof(axis));
    if (vxGetStatus((vx_reference)context) == VX_SUCCESS && vxGetStatus((vx_reference)s_axis) == VX_SUCCESS) {
        vx_reference params[] = {
            (vx_reference) input,
            (vx_reference) indices,
            (vx_reference) output,
            (vx_reference) s_axis,
        };
        node = createNode(graph, VX_KERNEL_GATHER_LAYER_AMD, params, sizeof(params) / sizeof(params[0]));
        vxReleaseScalar(&s_axis);
    }

    return node;
}